|
[Sponsors] |
April 9, 2016, 15:18 |
Finite Volume Method
|
#1 |
Member
Join Date: Apr 2016
Posts: 30
Rep Power: 10 |
Hi,
Could someone help with a little problem that i have regarding how FVM is applied in CFD. After discretization,we assume a general function that can be first order upwind,second order upwind etc.How to decide which kind of function to take? And also what are the effects of larger cell size in FVM? |
|
April 10, 2016, 09:43 |
|
#2 |
Senior Member
Michael Prinkey
Join Date: Mar 2009
Location: Pittsburgh PA
Posts: 363
Rep Power: 25 |
First order upwind assumes that the solution is piecewise constant over the cells...the value at the each face is the same as value at the centroid. Second order upwind assumes a linear expansion with the centroid value being the cell-averaged value and the linear terms as the gradient of the function dotted with the vector connecting the cell centroid to the test location:
phi(x) = phi_c + grad(phi) . (x - x_c) Formal order of accuracy is important, but really, the use of higher order schemes, especially for advection/convection terms is governed by artificial diffusion. First-order upwind schemes are very diffusive, so you lose details very quickly in practical simulations. Errors scale in FVM as they do in finite difference, finite element, or even plain-old Talyor series expansions. Errors increase as you make FV larger. The rate of increase is governed by the order of the method. First-order schemes scale error linearly. Second-order sceheme scale error quadratically, etc. |
|
April 11, 2016, 05:41 |
|
#3 | |
Senior Member
Filippo Maria Denaro
Join Date: Jul 2010
Posts: 6,849
Rep Power: 73 |
Quote:
The FV is based on the integral form of the equation. The function you are referring is the numerical flux function. Upwind (first, second or higher) flux has intrinsically numerical viscosity (mprinkey already wrote) while central flux discretizations have not. What you use depends on the flow case and formulation. DNS/LES use discretization without numerical viscosity conversely RANS use also upwind. |
||
April 11, 2016, 05:48 |
upwind scheme for Euler equation
|
#4 |
Member
sajad
Join Date: Apr 2014
Location: Iran
Posts: 46
Rep Power: 12 |
Does upwind scheme cause problem for Euler equation?
|
|
April 11, 2016, 05:54 |
|
#5 |
Senior Member
Filippo Maria Denaro
Join Date: Jul 2010
Posts: 6,849
Rep Power: 73 |
||
May 21, 2016, 03:28 |
|
#6 | |
Member
LUQILIN
Join Date: May 2016
Location: Harbin, China
Posts: 35
Rep Power: 10 |
Quote:
What you say about upwind is the way to compute flux at cell interface when discrete convention terms. 2. You can test the effects of cell size yourself. But if time integration is an explicit scheme, the cell size is not independent, i.e., associated with time steps. Check CFL number and ensure that your scheme is stable. |
||
Tags |
discretization, function, fvm, node, size |
|
|
Similar Threads | ||||
Thread | Thread Starter | Forum | Replies | Last Post |
[blockMesh] error message with modeling a cube with a hold at the center | hsingtzu | OpenFOAM Meshing & Mesh Conversion | 2 | March 14, 2012 10:56 |
On the damBreak4phaseFine cases | paean | OpenFOAM Running, Solving & CFD | 0 | November 14, 2008 22:14 |
[blockMesh] Axisymmetrical mesh | Rasmus Gjesing (Gjesing) | OpenFOAM Meshing & Mesh Conversion | 10 | April 2, 2007 15:00 |
ALE in finite volume method | littlelz | Main CFD Forum | 5 | June 21, 2003 13:50 |
finite volume method for CFD analysis of 2D blunt body | Aditya Vaze | Main CFD Forum | 1 | January 19, 2000 14:55 |